In this paper, the interaction patterns and their automatic analysis in spontaneous spoken information-seeking dialogues are studied. First, based on previous work from discourse analysis (i.e., exchange as basic interaction unit in Birmingham School) and Sytemic Functional Grammar (i.e., Halliday’s speech function), a principled scheme is proposed to model interaction patterns with utterance groups. Then a dialogue corpus is annotated with this scheme and further analyzed. Some main factors affecting the structure of utterance group are distinguished. Based on these, an algorithm is established to analyze utterance groups and is evaluated in the corpus. The results achieve a correct rate of 55.4%～84.2% for overall utterance tags, depending on the different recognition performances of the extended sentence type and utterance topic.